NVIDIA releases Nemotron 3.5 Lightning, a hybrid open-weight model with strong coding benchmarks

NVIDIA dropped Nemotron 3.5 Lightning (30B-A3B, hybrid Mamba+MoE+Attention, 1M context, open weights) with strong SWE-bench (52.8%) and PinchBench (85.4%) scores. Multi-token prediction and GRPO RL are notable. This open-source flagship model targets code/tool tasks and is available with weights.